We explore the practical relevance from a supervisor's viewpoint of a recent but already popular market-based indicator of the systemic importance of financial institutions, the marginal expected shortfall (MES). The MES of an institution can be defined as its expected equity loss when the market itself is in its left tail. We compute the dynamic MES developed by Brownlees and Engle (2010) for a panel of 65 large US banks over the last decade and a half. Running panel regressions of the MES on bank characteristics, we first find that the MES can be partly rationalized in terms of standard balance sheet indicators of bank health and systemic importance, but also that these relationships changed widely over time. We then ask whether the cross section of the MES can help to identify ex ante, i.e. before a crisis unfolds, which institutions are the more likely to suffer the most severe losses ex post, i.e. once it has unfolded. Unfortunately, using the recent crisis as a natural experiment, we find that standard balance-sheet metrics like the tier one solvency ratio are better able to predict equity losses conditional to a true crisis.